PEST is an advanced software tool for model calibration, parameter estimation and predictive uncertainty analysis. It is open-source and freely available, and currently distributed by S. S. Papadopoulos & Associates, Inc.
You will need to install PEST separately on your computer, and ensure that it can be executed from the command line. How to install PEST
PEST communicates with Simile models by creating files containing parameter values, running the model then reading files containing model output values. The model will be run many times. How to set up PEST interaction
When using PEST, model inputs and outputs can be individual values or sequences over time. If an output is a time series, the series itself can be plotted on a display tool alongside the model output. Running PEST
Predictive analysis can be used to get an estimate of a model output at a time later than that covered by available data, or it can be used to provide an uncertainty band around a model output trace. Working with predictive analysis
In: Contents >> Running models
PEST is an advanced software tool for model calibration, parameter estimation and predictive uncertainty analysis. It is open-source and freely available, and currently distributed by S. S. Papadopoulos & Associates, Inc. The Simile interface has been verified to work with PEST version 9 and later, on Windows, Linux and MacOS.
Windows users can download a pre-compiled version of PEST from this site and follow the instructions to enable it to run from the command line. Typing 'pest' to the command prompt will produce a message about usage if it is correctly installed.
Linux and Mac users will need to compile PEST from source. PEST is written in FORTRAN, and the GNU g95 compiler is suitable for compiling it. This is part of the GNU compiler collection, and is freely available from SourceForge. Some Linux distributions have the gfortran extension to gcc available from their software repositories, which is also suitable for compiling PEST.
It should be possible to build the package using the makefile shipped with it. The simplest way to get this to work is to create a symlink called 'f90' to the g95 compiler, since the makefiles all include f90 as the default compiler name. This can be placed in a directory in the executable search path, e.g., ~/bin. A symlink to the PEST executable can also be made in this directory once it is compiled.
Check that PEST is installed correctly by typing 'pest' to the command prompt before attempting to run it from within Simile. If it is set up correctly, it will generate a usage help message.
In: Contents >> Running models >> PEST
Once you have got your model behaving in a generally believable way, you can use PEST to calibrate your model parameters in order to match the model's output to a set of measured data.
Details of the operation of PEST can be found in its manual. The manual is included in the pre-compiled Windows distribution. Only the basics of PEST operation will be covered here.
The PEST interface is included in Simile as a helper. It is located in the 'Standard tools' submenu of the helpers menu. When you select it, it will create a tabbed notebook with four pages inside its top-level window. These are labelled Inputs, Outputs, Actions and Settings. If using the single-window Model Run Environment, be careful not to confuse these notebook tabs with those containing other helper pages.
On this page you specify which model parameters will be estimated by PEST. You can add a single parameter by hitting the '+' button at the top, then clicking on the parameter in the model diagram or explorer tool. The '-' button allows you to remove a parameter. The last button adds all suitable parameters that have not already been added. Note that PEST can only optimize real (floating-point) value parameters.
A minimum and maximum value must be provided for each parameter. For fixed parameters you can type these into entry boxes. They must reflect the range of values which are plausible. For variable parameters, PEST uses the values entered in the parameter's properties before the model was built.
For variable parameters you can decide whether PEST simply picks a fixed value for them or generates a series of values. There is a checkbox and an entry field for your choice. A series will be spaced regularly over the length of the model run, which is specified in the Actions page. Note that the total number of values being estimated (including all members of each array and each series) must not exceed the number of data measurements being used, or PEST will not have enough information to estimate them all.
PEST also needs an initial estimate for each parameter value. This defaults to midway between the parameter's minimum and maximum value, but you can enter a different estimate. If you select the 'Use current data as estimates' checkbutton at the bottom of the Inputs page, PEST will use the current values at the start of the optimization for its initial estimates.
If an array parameter is used, PEST will use the same specification data for each array member, but will create a separate estimate for each member.
On this page you specify which model values are to be looked at by PEST and compared with measured values. Any value in the model can be used, except values from inside variable-membership submodels. You can add a value by hitting the '+' button at the top, then clicking on the value in the model diagram or explorer tool. This indicates a component whose value at the end of each model run is to be compared with measured data. You can also add a value by hitting the graph icon button. This indicates a component whose value is to be sampled at a series of time points during a model run, and each sample compared with measured data. The '-' button allows you to remove a value of either type.
For each value you also need to load measurement data for comparison. PEST displays an input tool similar to the one in the 'File Parameters' dialogue, and data can be entered directly or loaded from files exactly as described for this dialogue.
When you have loaded your measurement data you can save references to it in a scenario file, as you would for parameter data. This scenario file will have the extension .smf rather than .spf to indicate that it refers to measurement data rather than parameter data. In other respects the two file formats are interchangeable.
Simile is a much mode expressive modelling tool than the typical command-line-driven models with which PEST is intended to work, so there are some limits on what Simile model constructs can be used when estimating parameters with PEST. In particular,
In: Contents >> Running models >> PEST
When you have set up the inputs and outputs in your model that will be used by PEST, you can go the the Actions tab in the PEST interface tool notebook and try to run it.
The run length from the run control will be used unless you alter it in the input box. If you do not select the checkbox in the Predictive analysis section, a simple parameter estimation will be done. Note that if all your outputs are time series, the run length will be up to the last time series value and the entry box for it will have no effect.
A PEST run can be controlled in the same way as a model run
While PEST is running you will see the run control values and progress bar update during each run. You will also see statistics about the run appear in the execution monitor box of the Actions tab. The text box in the execution monitor displays PEST's output to the command line, which is useful in case it should fail for any reason.
Simile's other helper tools are also updated while PEST runs. This means that if you have a slider tool for the variable parameters which are under PEST's control, you can watch them update as PEST adjusts them to optimize the outputs. However you may not want the tools to be updated, as they may slow down execution, and the plotter will show a separate trace for each run. To turn off all helper updates, set the 'Display interval' entry in the Run Control dialogue to zero. Alternatively you can set it to an interval longer than the run length, so the helpers (including sliders) will only be updated at the start of each run.
When PEST has finished optimizing, the model's parameters will be set to the values it has estimated. You can now go to the file parameters dialogue and save these values in a scenario file if they will be required later. PEST provides an indication (the 'phi' value) of how good its estimation is, and this is reported in the Actio tab of the helper -- the lower the phi value, the better the estimation. For a graphical view, if you have a time series of measurements and you are displaying the corresponding model value on a tool such as the plotter, you can select the 'show these values on plots' checkbox at the bottom of the PEST tool Outputs panel, then reset and run the model again, and a trace of the measured data will be superposed on the model data trace in the plotter tool.
PEST produces a selection of output files during a run. These are kept in Simile's temporary directory. When PEST has finished running, you can save any of these in a permanent directory using the 'Save a PEST file...' button. Each pest file has a different extension; use the extension chooser in the save dialogue to choose the file you want, and enter the name you want to give it.
If PEST fails to converge on a good set of parameters, you can adjust its settings. The global settings, such as NOPTMAX to control the maximum allowed number of optimization stages, are on the Settings tab. Settings related to each input or output, such as PARCHGLIM which affects how the maximum change of an input value on a single step is applied, can be edited in a context menu for that input or output
In: Contents >> Running models >> PEST
When you first run PEST, it runs in parameter estimation mode. This selects values for the input parameters which cause the outputs to most closely follow the measured values, according to the least-squares rule. The quality of fit is given by PHI
The purpose of predictive analysis is to determine the range over which a model output can vary while still being consistent with the measured values supplied. To get maximum and minimum predictions, PEST allows the PHI value to be larger than the minimum value found during parameter estimation.
To do a predictive analysis on a Simile model, check the checkbox in the predictive analysis frame on the Actions tab, and select a model value for which to make a prediction. This can be any non-parameter value in the model, not including those in variable-membership submodels. It can be a value which you are already using for comparison with measured data
You can also make the following selections:
Whether to predict maximum or minimum values
What multiple of the original PHI to allow for the predictions
At what time(s) the prediction will apply.
There are some values on the Settings tab which affect PEST's behaviour during predictive analysis; these are explained in the PEST manual. Simile supplies default values for these, based on the results obtained when running PEST in parameter estimation mode.
If the value being predicted is an array, or if the frequency and duration you set for prediction times are such that a prediction will be made at more than one time, then a complete run of PEST must be done for each prediction, because PEST itself can only predict one value at a time. Since each PEST run includes multiple model runs, predicting large numbers of values can take a very long time. The following steps might speed things up:
Set display interval to 0, so display tools do not take up time
Select 'use current values as estimates' on the Inputs tab after the parameter estimation run, so each prediction run starts with good parameter estimates.
During a prediction run, the progress bar shows how many of the prediction runs have been completed. Predicted values will be inserted into the results field, and can be displayed in tabular form and saved as a .csv file using the View button.
It is worth noting that each prediction is arrived at independently. So, if using predictive analysis to create 'error bands' around a graph of model outputs, it might be the case that although each individual value might vary as far as the predicted limits, it would be impossible for the graph line to actually follow the minimum or maximum limits, since multiple predicted values on the limit might be incompatible with each other. Remember, however clever PEST might appear, it's still only a computer program.
In: Contents >> Running models >> PEST